Highlights
🎭 Skill-Embedded MCP Support
Skills can now embed MCP server configurations directly in their YAML frontmatter or via mcp.json files. This enables skills to bring their own tooling - load a skill, get its MCP capabilities automatically.
🎪 Built-in Playwright Skill
New playwright skill ships out of the box for browser automation, web scraping, testing, and screenshots. Disable via disabled_skills: ["playwright"].
📁 Recursive Command Loader
Commands now load from subdirectories recursively. Organize your .claude/commands/ with nested folders.
⚡ Recovery Pipeline Overhaul
Changed phase order to DCP → Truncate → Summarize for more aggressive context recovery when hitting token limits.
Changelog
- 13ebeb9 fix: correct preemptive_compaction schema comment default value (#400)
- 2452a47 fix(tests): resolve mock.module leakage breaking ralph-loop tests
- 44640b9 docs(agents): update AGENTS.md with skill-mcp feature documentation
- 794b526 fix(keyword-detector): remove word boundary requirement for ulw/ultrawork detection
- b0c39e2 feat(builtin-skills): add playwright skill with MCP config and disabled_skills option
- bd05f5b feat(skill_mcp): add dynamic truncation and grep filtering
- a82575b feat(skill): display MCP tool inputSchema when loading skills
- ff760e5 feat(skill-loader): support mcp.json file for AmpCode compatibility
- 4039722 feat(plugin): integrate skill_mcp tool with session-scoped lifecycle management
- 439785e feat(skill): display MCP server capabilities when skill with MCP is loaded
- e533031 feat(tools): add skill_mcp tool for invoking skill-embedded MCP operations
- b122273 feat(skill-loader): parse MCP server config from skill frontmatter
- 06dee72 feat(skill-mcp): add MCP client manager with lazy loading and session cleanup
- c8aed3f chore(deps): add @modelcontextprotocol/sdk and js-yaml for skill-embedded MCP support
- 1d4b5de feat(mcp): restrict grep_app tools to librarian agent only (#395)
- a217610 Fix Bun mock.module() leak between test files preventing ralph-loop tests from passing
- b377571 Update AGENTS.md documentation hierarchy with auth and hooks details
- 490c0b6 Add auto-slash-command hook for intercepting and replacing slash commands
- b30c17a fix(recovery): more aggressive truncation, remove revert fallback
- a5983f1 fix(anthropic-context-window-limit-recovery): add revert fallback when truncation insufficient
- 2948d94 Change recovery phase ordering to DCP → Truncate → Summarize
- c66cfbb Remove invalid model reference from publish command
- f66c886 feat(keyword-detector): show toast notification when ultrawork mode is activated (#393)
- 1c55385 feat(command-loader): add recursive subdirectory scanning for commands (#378)
- f3db564 fix: reduce context duplication from ~22k to ~11k tokens (#383)
- 15b0ee8 feat(doctor): add GitHub CLI check (#384)
Thank you to 2 community contributors:
- @sisyphus-dev-ai:
- feat(doctor): add GitHub CLI check (#384)
- fix: reduce context duplication from ~22k to ~11k tokens (#383)
- feat(keyword-detector): show toast notification when ultrawork mode is activated (#393)
- feat(mcp): restrict grep_app tools to librarian agent only (#395)
- fix: correct preemptive_compaction schema comment default value (#400)
- @ul8:
- feat(command-loader): add recursive subdirectory scanning for commands (#378)